Predictors of the number of under-five malnourished children in Bangladesh: application of the generalized poisson regression model

Islam et al. BMC Public Health 2013, 13:11http://www.biomedcentral.com/1471-2458/13/11

RESEARCH ARTICLE Open Access

Predictors of the number of under-fivemalnourished children in Bangladesh: applicationof the generalized poisson regression modelMohammad Mafijul Islam1,2, Morshed Alam1, Md Tariquzaman3, Mohammad Alamgir Kabir1,4, Rokhsona Pervin5,Munni Begum6 and Md Mobarak Hossain Khan7*

Abstract

Background: Malnutrition is one of the principal causes of child mortality in developing countries includingBangladesh. According to our knowledge, most of the available studies, that addressed the issue of malnutritionamong under-five children, considered the categorical (dichotomous/polychotomous) outcome variables andapplied logistic regression (binary/multinomial) to find their predictors. In this study malnutrition variable(i.e. outcome) is defined as the number of under-five malnourished children in a family, which is a non-negativecount variable. The purposes of the study are (i) to demonstrate the applicability of the generalized Poissonregression (GPR) model as an alternative of other statistical methods and (ii) to find some predictors of thisoutcome variable.

Methods: The data is extracted from the Bangladesh Demographic and Health Survey (BDHS) 2007. Briefly, this surveyemploys a nationally representative sample which is based on a two-stage stratified sample of households. A total of4,460 under-five children is analysed using various statistical techniques namely Chi-square test and GPR model.

Results: The GPR model (as compared to the standard Poisson regression and negative Binomial regression)is found to be justified to study the above-mentioned outcome variable because of its under-dispersion(variance < mean) property. Our study also identify several significant predictors of the outcome variablenamely mother’s education, father’s education, wealth index, sanitation status, source of drinking water, andtotal number of children ever born to a woman.

Conclusions: Consistencies of our findings in light of many other studies suggest that the GPR model is an idealalternative of other statistical models to analyse the number of under-five malnourished children in a family. Strategiesbased on significant predictors may improve the nutritional status of children in Bangladesh.

Keywords: Malnutrition, Under-five children, Predictors, Generalized Poisson regression model, Bangladesh

BackgroundMalnutrition among children is a major public healthproblem in developing countries including Bangladesh[1-10]. Globally children with moderate and severe acutemalnutrition are approximately 60 million and 13 millionrespectively [2]. Between 8 and 11 million under-fivechildren also die each year in the world [2,9]. More than50% of these deaths are attributed to malnutrition, which

* Correspondence: [email protected] of Public Health Medicine, School of Public Health, Universityof Bielefeld, Bielefeld, GermanyFull list of author information is available at the end of the article

© 2013 Islam et al.; licensee BioMed Central LCommons Attribution License (http://creativecreproduction in any medium, provided the or

are mostly preventable through economic developmentand public health measures [2]. Although Bangladesh hasalready achieved a remarkable progress in reducing childmalnutrition from 68% in the late 1980s to 41% in 2007[6,11] and under-five mortality [12], still malnutrition is acommon problem in this country [13,14]. It is one of thecountries with very high burden of malnutrition [5,14].The underlying cause for 60% of the under-five deaths ismalnutrition in Bangladesh [5].Malnutrition among children is a critical problem be-

cause its effects are long lasting and go beyond child-hood. It has both short- and long-term consequences

td. This is an Open Access article distributed under the terms of the Creativeommons.org/licenses/by/2.0), which permits unrestricted use, distribution, andiginal work is properly cited.

Islam et al. BMC Public Health 2013, 13:11 Page 2 of 8http://www.biomedcentral.com/1471-2458/13/11

[6,8,10]. For instance, malnourished as compared tonon-malnourished children are physically, emotionallyand intellectually less productive and suffer more fromchronic illnesses and disabilities [6,15,16]. Malnutritionamong children depends on complex interactions ofvarious factors reflecting socio-demographic, environ-mental, reproductive, institutional, cultural, political andregional factors [3,4,6,13,14,17-19]. Already many studieshave been conducted to find the predictors of malnutri-tion in Bangladesh and elsewhere [7,8,20-28].Poverty is found to be strongly associated with child

malnutrition [4]. Although the relationship betweeneconomic ability and malnutrition is complex, a num-ber of studies have illustrated that children of poorerhouseholds tend to be more undernourished thanchildren of wealthier ones [3,13,22,24-26,29]. Parentaleducation is also identified as a strong precdictor ofchild malnutrition [14,28]. However, the association ofmaternal education is relatively stronger than parentaleducation [14,20,29]. Other determinants of childmalnutrition may include social deprivation [23],rural-urban place of residence [18,30], religions[31,32], number of children [1,18], source of drinkingwater [1] and toilet facility [3,33].Most of the abovementioned studies, that addressed

the topics of child malnutrition, used categorical out-come variables and applied logistic regression (multi-nomial/binary) [7,10,14,27-29,34] models to find thepredictors of child malnutrition. For instance, Mashalet al. [27] used multivariable logistic regression to iden-tify the risk factors of malnutrition among children inAfghanistan. Mueller et al. [34] applied logistic regres-sion to study the relationship of malnutrition with mor-bidity and mortality among West African children.Ojiako et al (2009) used Tobit model to find thedeterminants of malnutrition among preschool childrenin Nigeria [19]. Sometimes we see studies whichaddressed non-negative outcome variables like numberof children in a household [35,36] and number ofaccidents [37]. Recently different types of regressionbased on Poisson distribution namely standard Poissonregression model, negative binomial regression modeland generalized Poisson regression (GPR) model havebeen used to model such kind of count variables [36-38].However, applications of these models are based on cer-tain assumptions. For instance, standard Poisson regres-sion model assumes equal mean and variance of thedependent variable. In reality, often this equality as-sumption is not true because the variance could behigher than mean (over-dispersion property) or lowerthan mean (under-dispersion property). Ignorance ofthese properties may produce biased standard errors andinefficient estimates of regression parameters, althoughthe estimates of the standard Poisson model are still

consistent. The negative binomial regression model ismore flexible than the standard Poisson model and isfrequently used to analyse outcome variable with over-dispersion. The GPR model, on the other hand, can cap-ture both over- and under-dispersion properties of thedata, which make this model even more flexible [37].However, a count variable like number of children ina household often shows under-dispersion property[35]. Our data also shows the under-dispersion prop-erty. Therefore, we applied the GPR model to ourdata. The first objective was to demonstrate the ap-plicability of this model as an alternative to study thechild malnutrition in Bangladesh. The second object-ive was to find some predictors of the count variabledefined as the number of under-five malnourishedchildren in a Bangladeshi family.

MethodsData sourceThe data was extracted from the Bangladesh Demo-graphic and Health Survey (BDHS) conducted in 2007.This survey employs a nationally representative samplewhich is based on a two-stage stratified sample ofhouseholds. This type of survey generally provides infor-mation on basic national indicators of social develop-ment. The BDHS 2007 was a part of the globalDemographic and Health Survey (DHS) programme[39]. The present study utilised the information of 4,460under-five children aged 0 to 59 months for whom an-thropometric measurements were available.

Dependent variableResearchers can define nutritional status of children dif-ferently. The nutritional status of a child is typicallybased on several measurements namely height, weight,sex and age of the child. Three commonly usedmeasures for nutritional status are height-for-age,weight-for-height and weight-for-age [40]. Thesemeasures are then expressed as Z-scores from the me-dian of the reference population. In our study, we use‘weight-for-height’ because it can describe current nutri-tional status by linking body mass in relation to bodylength. It does not require the exact age information ofthe child, which is necessary for the ‘weight-for-age’[39]. It can also track whether a child recently receivessufficient contents of nutrients to build and maintainbodyweight along with other factors such as geneticgrowth, environment, and disease burden on activitylevel [41]. To define child malnutrition, we followed thenational report of Bangladesh [39] and the guidelines ofthe World Health Organization [41]. According to thesereports, a child is malnourished if the Z-score is below -2 standard deviation (SD) from the median of thereference population. The dependent variable in this

Islam et al. BMC Public Health 2013, 13:11 Page 3 of 8http://www.biomedcentral.com/1471-2458/13/11

study was expressed as the number of under-five mal-nourished children in a Bangladeshi family.

Covariates/predictor variablesWe consider several covariates (Table 1) as predictorswhich are commonly reported in the nutritional studiesof children. Some of them (categories are given in paren-theses) are place of residence (urban, rural ), parentaleducation (no education, 1-5 years education, 6-10 yearseducation, 11+ years education), father’s occupation(professional, business, farmer, worker; where workermeans skilled, semi-skilled, factory worker, and bluecolor service), toilet facility (yes or no), sources ofdrinking water (tube well water, piped water, others;where others mean dug well water, unprotected water,river or dam or lake or ponds or canal water, rainfallwater), religion (Islam, others; where other religions in-clude Hinduism, Buddhism, Christian, and unknownreligions), access to media (yes, no), wealth index (lowest,second, middle, fourth, highest), total number of childrenever born to a woman, and total number of the childrendied in a family. Two of these variables namely wealthindex and access to media are composite variables. Thewealth index is an asset-based index that reflects the rela-tive socioeconomic status of the household and is widelyused in low- and middle-income countries to quantify in-equalities and to control the confounding effect of

Table 1 Basic characteristics of the parents and households in

Predictors n (%) Mean Predictors

Place of residence: Sources of

Urban 1590 (35.70) Piped wate

Rural 2870 (64.30) Tube well w

Mother’s education: Others

No education 1136 (25.50) Religion:

1-5 years education 1371 (30.70) Islam

6-10 years education 1566 (35.10) Others

11+ years education 385 (8.60) Access to m

Father’s education: Yes

No education 1437 (32.20) No

1-5 years education 1240 (27.80) Wealth ind

6-10 years education 1184 (26.50) Lowest qui

11+ years education 594 (13.30) Second qu

Father’s occupation: Middle qui

Farmer 1118 (25.53 Fourth quin

Worker 2003 (45.75) Highest qu

Professional 201 (4.59)

Business 1056 (24.12) Total numb

Toilet facility: Total numb

Yes 3286 (74.00)

No 1153 (26.00)

socioeconomic variables. It is based on the household own-ership variables (e.g. car, refrigerator, television), housingcharacteristics (e.g. materials of the floor, roof, walls) andaccess to services (e.g. availability of electricity) [12]. Accessto media, which is also a composite index, is based onthree mass media variables namely whether they listen toratio, watch television, and read newspaper or magazine.This is categorized into two groups, where ‘yes’ meansrespondents have access to at least one of these media and‘no’ means no access to any of these media.

GPR modelThe generalized Poisson probability function of thenumber of malnourished children (Y) in a family can bewritten as

f y; μ; αð Þ ¼ μ

1þ αμ

� �y 1þ αyð Þy�1

y!exp � 1þ αyð Þ

1þ αμð Þ� �

;

y ¼ 0; 1; 2; . . .

ð1ÞThe mean and variance of Y are given by E(Yi| xi) = μi

and V(Yi| xi) = μi(1 + αμi)2, where the mean of the

dependent variable is related to the explanatory variablesthrough the link function μi = μi(xi) = exp(xiβ). In this linkfunction, xi is a k - 1 dimensional vector of covariates, β isa k-dimensional vector of regression parameters, and α is a

Bangladesh based on BDHS 2007

n (%) Mean

drinking water:

r 299 (6.70)

ater 3595 (80.60)

566 (12.70)

4050 (90.81)

410 (9.19)

edia:

919 (29.10)

3127 (70.10)

ex:

ntile 849 (19.00)

intile 901 (20.20)

ntile 835 (18.80)

tile 850 (19.00)

intile 1025 (23.00)

er of children ever born to a woman 2.67

er of children dead in a family 0.24

Table 2 Bivariate associations between the number ofunder-five malnourished children in a family anddifferent predictors in Bangladesh, 2007

Characteristics χ2 P

Place of residence 55.36* <0.001

Mother’s education 252.70* <0.001

Father’s education 241.91* <0.001

Father’s occupation 123.75* <0.001

Wealth index 253.21* <0.001

Sources of drinking water 34.45* <0.001

Toilet facility 47.48* <0.001

Religion 0.01 0.925

Access to media 14.49* <0.001

Total number of children ever-born to a woman 79.55* <0.001

Total number of children died in a family 30.78* <0.001

Islam et al. BMC Public Health 2013, 13:11 Page 4 of 8http://www.biomedcentral.com/1471-2458/13/11

dispersion parameter. The standard Poisson regressionmodel is a special form of the generalized Poisson regres-sion model. When α is equal to zero, the probability func-tion of generalized Poisson random variable reduces to thePoisson probability function. The positive value of α inequation (i) indicates the over-dispersion, whereas thenegative value of α indicates the under-dispersion propertyof the distribution.For selecting the right type of Poisson regression

model, it is necessary to check the existence of disper-sion problem in the data. The moment estimators of thetwo parameters in the Poisson distribution given byConsul and Jain [42] are as follows:

μ̂ ¼ffiffiffiffiffi�y3

s2

r

And

α ¼ 1�ffiffiffiffi�ys2

r

Where ӯ and s2 are sample mean and variance respect-ively. The asymptotic variances of the moment estimatorsgiven by Shoukri [43] are:

V μ̂ð Þ≈ μ̂

2nμ̂ þ 2� 2α̂ þ 3α̂2

1� α̂

� �;

And

V α̂ð Þ≈1� α̂

2nμ̂μ̂ � μ̂α̂ þ 2α̂ þ 3μ̂2� �

:

The adequacy of the GPR model over the PR model isassessed by setting the following hypothesis

H0 : α ¼ 0

Versus

H1 : α≠0:

This test of hypothesis determines whether the disper-sion parameter is statistically different from zero. Therejection of H0 recommends the use of the GPR modelrather than the standard Poisson regression model. Toperform the test, the asymptotically normal Wald type“Z” statistic defined as the ratio of the estimate of α toits standard error is used.The estimation of regression coefficients β is obtained

by the maximum likelihood approach. The log-likelihood functions of the GPR model is

Log L β; α; yð Þð Þ ¼Xn

i¼1½yi log μi

1þ αμi

� �þ yi � 1ð Þ log 1þ αyið Þ

� μi 1þ αyið Þ1þ αμi

� log y!ð Þ�

ð2Þ

Where

μi ¼ μi xið Þ ¼ exp xiβð Þ

Statistical analysisSimple summary statistics (either as percentage for the cat-egorical variables or mean for the continuous variables) areshown for selected socioeconomic predictors (Table 1). Atthe outset of analyses, sample mean and sample varianceof the dependent variable are calculated in order to checkwhether it follows the standard Poisson regression modelor GPR model. Then the Z test is performed to checkwhether the dispersion parameter significantly deviatesfrom zero. Here the null hypothesis (H0 : α = 0) states thatthe value of dispersion parameter is zero. In contrast, atwo-sided alternative hypothesis (H1) is used whichindicates that the value of the dispersion parameter is un-equal to zero. Bivariate analyses (based on Pearson Chi-square test) are performed to examine association betweendependent variable and each of the selected predictors(Table 2). All significant predictors are then finally includedinto the GPR model. As the dependent variable is more ap-propriate for the GPR model because of its under-dispersion property, we applied this model to estimate theregression parameters (β) including ‘p’ values based onWald Chi-square values. Finally, incidence rate ratio (IRR)and 95% confidence interval are calculated for each groupof the categorical predictors (Table 3). The statistical soft-ware packages SAS 9 and SPSS 11.5 are used to extract theinformation from BDHS 2007, to recode the variables, andto perform univariate and bivariate analyses. Finally weused R 2.14.1 to estimate parameters of the GPR model.

ResultsThe estimated mean �Y ¼ 0:626ð Þ and variance (sy

2 = 0.369)of the outcome variable reveal the under-dispersion

Table 3 Results of the multivariable generalized Poisson regression analysis to study the number of under-fivemalnourished children in Bangladesh, 2007

Predictors Categories Estimated regressioncoefficient (ß)

χ2 P-value Estimated IRR 95% CI for IRR

Place of residence: Urban 0.06 0.44 0.507 1.07 0.88-1.28

Rural(r)

Mother’s education: No education 0.33* 7.01 0.008 1.39 1.09-1.78

1-5 years education 0.32* 6.81 0.009 1.37 1.08-1.74

6- 10 years education 0.22 3.00 0.083 1.24 0.97-1.59

11+ years education(r)

Father’s education: No education 0.29* 5.17 0.023 1.33 1.04-1.71

1-5 years education 0.26* 3.92 0.048 1.30 1.00-1.68

6- 10 years education 0.20 2.26 0.133 1.22 0.94-1.58

11+ years education(r)

Father’s occupation: Farmer 0.39 2.45 0.118 1.48 0.91-2.43

Worker 0.32 2.04 0.153 1.39 0.89-2.17

Professional 0.11 0.14 0.705 1.12 0.63-1.97

Business(r)

Wealth index: Lowest quintile 0.50* 22.90 <0.001 1.64 1.34-2.01

Second quintile 0.41* 10.38 0.001 1.50 1.17-1.93

Middle quintile 0.33* 12.46 <0.001 1.39 1.16-1.67

Fourth quintile 0.22* 4.46 0.035 1.25 1.02-1.53

Highest quintile(r)

Sources of drinking water: Piped water −0.24 3.18 0.075 0.79 0.61-1.02

Tubewell water −0.34* 16.48 <0.001 0.71 0.60-0.84

Others(r)

Toilet facility: No 0.36* 56.32 <0.001 1.43 1.30-1.56

Yes(r)

Access to media: No 0.08 1.41 0.235 1.08 0.95-1.22

Yes(r)

Total number of children ever born to a woman 0.05* 11.04 0.001 1.06 1.02-1.09

Total number of children dead in a family −0.03 0.16 0.688 0.97 0.83-1.13

Notes: (r) indicates the reference group in each category.* p < 0.05.

Islam et al. BMC Public Health 2013, 13:11 Page 5 of 8http://www.biomedcentral.com/1471-2458/13/11

property of the data. In the total sample, 16.9 percent ofthe under-five children are malnourished. Table 1 providesdescriptive statistics for all predictors. According to thistable, illiteracy rate is lower among mothers of children(26 percent) as compared to their fathers (32 percent).In contrast, the rate of higher education is higheramong fathers of children (13.3 percent) than theirmothers (8.6 percent). Nearly 30 percent of the familieshave access to the mass media. About one-fourth of thefamilies (26 percent) have no toilet facility. The meannumber of children ever born to a woman is 2.67.Table 2 shows the summary results of bivariate ana-

lyses between outcome and predictor variables. All thepredictors except religion show significant associationswith outcome variable.The estimated value of the dispersion parameter (α)

and its standard deviation from equation (i) are -0.30266

and 0.000439, respectively. Our Null hypothesisformulated as H0 : α = 0, (Z=-14.4405 and p < 0.05) isrejected at 5% level of significance.According to the results of GPR analysis (Table 3),

variables namely mother’s education, father’s education,wealth index, toilet/sanitation, sources of drinking water,as well as total number of children are statisticallyassociated with the child malnutrition in a family. Theincidence rate of children suffering from malnutrition ishigher for mothers having no education (IRR=1.39; 95%CI=1.09-1.78) and 1-5 years education (IRR=1.37; 95%CI= 1.08-1.74) as compared to mothers with higher edu-cation. Similar results are also found for father’s educa-tion. The incidence rate of malnutrition among childrenis estimated to be nearly 1.64 times higher in the lowestquintile than the highest quintile. Children of middleand fourth quintiles also show higher incidence rate of

Islam et al. BMC Public Health 2013, 13:11 Page 6 of 8http://www.biomedcentral.com/1471-2458/13/11

malnutrition as compared to the children of highestquintile. The children who drink piped water andtubewell water are nearly 21% and 29% less likely to ex-perience malnutrition than the children drinking othersources of water (such as dug well water, unprotectedwell, surface water, unprotected spring, river or dam orlake or ponds or canal, rain water, etc). Similarly, toiletfacility is strongly associated with malnutrition status ofchildren. A child from a family having no sanitary toiletfacility has 1.43 times higher incidence rate of experien-cing malnutrition than a child with toilet facility. Thetotal number of children ever-born in a family and themalnourished children are also positively associated.

DiscussionOur study demonstrates that the GPR model is an idealalternative to study the malnutritional status of childrendefined as the number of under-five malnourished chil-dren in a family. This model is a good alternative be-cause most of the results of this study are found to beconsistent with the findings of many other studies[1,3,7,10,14,19,26,28,31,33,44,45].According to the GPR model, the malnutritional status

of children is insignificant between rural and urbanareas. This finding is both consistent [14] and contra-dictory [18]. A similar study in Vietnam [18] reportshigher level of malnutrition in rural areas than urbanareas. Some possible causes according to this study arelack of economic, socio-cultural, healthcare and intu-itional facilities in rural areas [18]. Like in Vietnam, ruralareas of Bangladesh also suffer from limited infrastruc-ture and facilities in terms of modern healthcareservices, sanitation, education, electricity and economicfacilities. Particularly health services are concentrated inurban areas than rural areas [12]. Although urban-ruraldisparity in terms of child malnutrition is negligible inBangladesh, this is not the case for many otherindicators. For instance, one recent study reports re-markable urban-urban disparities for antenatal care ser-vice, age at marriage, fertility and child mortality inBangladesh [46]. Another study in Vietnam also reportshigher age at marriage, smaller family size and lowermortality rate of children in urban areas as compared torural areas. What are the reasons for this inconsistency?One of the possible reasons might be related to the ad-justment of the GPR model by other potentialpredictors, namely household wealth and maternal edu-cation [30]. Inclusion of other predictors along withplace of residence into the same model may reduce thestrength of urban-rural nutritional disparity. However,further elaborative research is warranted in this regard.Many studies suggest that mother education is linked

with child health outcomes. The relationship of maternaleducation with child malnutrition is more demonstrable

than paternal education, health service availability, andsocioeconomic status [14,20,29,47]. However, some stud-ies also show parental educational effects on child nutri-tion [21,48]. Our study finds a significant positiverelationship between mother’s education and child nutri-tion. This result is consistent with many other studies[1,3,7,8,10,14,18,19,26,28,31,44,45,47,49]. Such a relation-ship could exist because maternal schooling is stronglyassociated with good child care and good health. Womenwith higher as compared to lower educational level aremore likely to raise their family income, which helps thefamilies to provide more quality diets and betterhealthcare to their children. Additionally, educatedmothers can efficiently use limited household resourcesand available healthcare facilities, limit their family size,maintain better health promoting behaviours and providehealthcare to their children [18,47]. All these factorspositively contribute to the child nutrition.The relationship between economic inequality and

children nutritional status is investigated by many stud-ies [3,26,29,45,50]. Generally the greater degree of eco-nomic disparity is associated with higher mortality [51].The relationship between economic disparity and mal-nutrition at the national level is not straightforward, be-cause better economy at the national level does notnecessarily mean better health care for all. Social andeconomic disparity in a country may differently influ-ence the accessibility to food and healthcare services in-cluding the burden of disease. A number of studies haveillustrated that children from poorer households aremore likely to be malnourished than children fromwealthier households [3,22,24,25]. Social deprivation isalso linked with a child's nutritional status [23]. InBangladesh the nutritional status of children differs indifferent economic classes [26], which can be attributedto the fact that rich families have more ability to allocatenecessary resources for their children than poor families.Understandably allocation of more resources to theirchildren improves their health conditions by reducingmultiple health risks.Our finding reveals a strong positive association between

number of children ever born to a woman and the numberof under-five malnourished children in a family. Theseresults are consistent with the findings of other study[1,18]. Generally families with more children experiencemore economic strain for food consumption and hencethey are more likely to suffer from poor nutritional status.In other words, inadequate allocation of householdresources among many children may lead to the low nutri-tional status. Particularly poor families cannot fulfil the nu-tritional requirements of the children. Families with morechildren generally devote less time to take care of theirchildren [18]. Because of negative impacts of higher fertilityon nutritional status of children, increasing birth interval

Islam et al. BMC Public Health 2013, 13:11 Page 7 of 8http://www.biomedcentral.com/1471-2458/13/11

should be an important strategy to improve the nutritionalstatus among the under-five children.In Bangladesh men are generally the main earner of a

family, although employment opportunities are increas-ing for women due to the flourishing garment sectors.Income of the family is strongly associated with the typeof father’s occupation. Normally fathers with more pres-tigious job have higher income than fathers with lowlevel jobs and therefore children from the higher incomefamilies should have better nutritional status. However,this is not the case in our study. The insignificant associ-ation of father’s occupation with the nutritional status ofchildren can be explained by the lack of proper nutri-tional knowledge and confounding effect of education ofparents. Like father’s occupation, religion does not playany significant role in explaining the nutritional status ofthe under-five children in Bangladesh.In our study factors namely source of drinking

water and type of toilet also show significant associ-ation with child malnutrition. Similar results arereported by Pongou et al [3]. These are plausible be-cause access to safe drinking water and hygienic toi-let are the pre-conditions for maintaining goodhygiene and nutrition among children. The incidenceof various water-borne illnesses can be reduced withthe improved supply of drinking water [15]. There-fore increasing access to the safe drinking water andhygienic sanitation are important to improve the nu-tritional status among under-five children.This study has several strengths. The use of nationally

representative data is one of the important strengths.Our findings could be reliable because of the large sam-ple. Application of the GPR model as an alternative ofother methods is another strength. Inclusion of rightpredictors into the model based on previous studies alsoincreases the strength of the study. However, this studyis not free from limitation. Firstly, all inherentlimitations associated with the cross-sectional data arealso true here. Another limitation of the study is that themodel does not include regional and cultural variables,which are also reported as significant predictors [13].Sex of the children is also not included in the model. Fi-nally some socioeconomic variables are stronglycorrelated with each other (e.g. wealth index and educa-tion), which may produce biased estimates because ofmulticollinearity [13]. Exclusion of important predictorsdue to non-availability may also alter our findings.

ConclusionsOur study demonstrates that the GPR model is an appro-priate model to identify predictors affecting the nutri-tional status of children in Bangladesh. Father’s andmother’s education, wealth index, source of drinkingwater of the household, toilet facility, and total number

of children ever born to a woman are significantlyassociated with child malnutrition in Bangladesh. Variousstrategies are reported by many studies [4-6,11,13,44]. In-creasing educational facilities for mothers and fatherscan improve the child nutrition. Facilitating access to safedrinking water and sanitation for poor families is also ne-cessary to improve the child nutrition. Since higher fertil-ity (i.e. number of children ever born to a woman) has anegative impact on child nutrition, government shouldimplement policies to limit family size by increasing birthspace [44]. Comprehensive and concerted nutritionalinterventions such as exclusive breastfeeding, comple-mentary feeding, supplementation of micronutrients tochildren and mother, hygiene interventions, and manage-ment of severe malnutrition are also needed to improvechild nutrition [4,8,11]. Other strategies such as publictransportation to carry food and relief programmes forthe disadvantaged groups are important to reduce childmalnutrition [13]. Addressing inequity and generaldeprivation and implementation of other healthprogrammes are also necessary to reduce malnutritionamong children [4]. We should keep in mind that ad-equate nutrition of children is a prerequisite to build ahealthy and productive nation [11]. In addition, toachieve the millennium development goal of halvingchild undernutrition by 2015, Bangladesh needs to scaleup target-oriented programmes such as poverty-reduction income generating interventions and improve-ment of public food transports for the poor populationand disadvantaged regions [13].

Competing interestsThe authors have no competing interests arising from the publication of thisarticle.

Authors’ contributionsMI, MA and MAK conceptualized the research topic together with MMHKand drafted the manuscript. MT mainly performed the data analysis. RPsignificantly contributed to the writing process and interpretation. MBrevised the article critically and provided further inputs. MMHK finallystructured the manuscript, collected the references and edited extensivelybefore finalization. All authors read and approved the final manuscript.

AcknowledgementsWe acknowledge support of the publication fee by DeutscheForschungsgemeinschaft and the Open Access Publication Funds of BielefeldUniversity.

Author details1Department of Statistics, Jahangirnagar University, Savar, Dhaka -1342,Bangladesh. 2Department of Mathematics & Statistics, Bowling Green StateUniversity, Bowling Green, OH 43402, USA. 3Probationary Senior Officer,Pubali Bank Ltd, Dhaka, Bangladesh. 4Department of Applied Statistics,University of Malaya, Kuala Lumpur, Malaysia. 5Department of AgriculturalStatistics, Sher-e Bangla Nagar Agricultural University, Sher-e Bangla Nagar,Dhaka -1207, Bangladesh. 6Department of Mathematical Sciences, Ball StateUniversity, Muncie, IN 47306, USA. 7Department of Public Health Medicine,School of Public Health, University of Bielefeld, Bielefeld, Germany.

Received: 23 August 2012 Accepted: 3 January 2013Published: 8 January 2013

Islam et al. BMC Public Health 2013, 13:11 Page 8 of 8http://www.biomedcentral.com/1471-2458/13/11

References1. Haughton D, Haughton J: Explaining child nutrition in Vietnam. Econ Dev

Cult Change 1997, 45(3):541–556.2. Collins S, Dent N, Binns P, Bahwere P, Sadler K, Hallam A: Management of

severe acute malnutrition in children. Lancet 2006, 368(9551):1992–2000.3. Pongou R, Ezzati M, Salomon JA: Household and community

socioeconomic and environmental determinants of child nutritionalstatus in Cameroon. BMC Publ Health 2006, 6:98.

4. Black RE, Allen LH, Bhutta ZA, Caulfield LE, de Onis M, Ezzati M, Mathers C,Rivera J: Maternal and child undernutrition: global and regionalexposures and health consequences. Lancet 2008, 371(9608):243–60.

5. Faruque ASG, Ahmed AMS, Ahmed T, Islam MM, Hossain MI, Roy SK, Alam N,Kabir I, Sack DA: Nutrition: basis for healthy children and mothers inBangladesh. J Health Popul Nutr 2008, 26(3):325–339.

6. Ahmed T, Ahmed AM: Reducing the burden of malnutrition inBangladesh. BMJ 2009, 339(b4490):1060.

7. Mohsena M, Mascie-Taylor CGN, Goto R: Association between socio-economic status and childhood undernutrition in Bangladesh; acomparison of possession score and poverty index. Publ Health Nutr 2010,13(10):1498–1504.

8. Jesmin A, Yamamoto SS, Malik AA, Haque MA: Prevalence and determinants ofchronic malnutrition among preschool children: a cross-sectional study inDhaka city, Bangladesh. J Health Popul Nutr 2011, 29(5):494–499.

9. Singer PA, Ansett S, Sagoe-Moses I: What could infant and young childnutrition learn from sweatshops? BMC Publ Health 2011, 11:276.

10. Abuya BA, Ciera JM, Kimani-Murage E: Effect of mother's education onchild's nutritional status in the slums of Nairobi. BMC Pediatr 2012, 12:80.

11. Ahmed T, Mahfuz M, Ireen S, Ahmed AMS, Rahman S, Islam MM, Alam N,Hossain MI, Rahman SMM, Ali MM, Choudhury FP, Cravioto A: Nutrition ofchildren and women in Bangladesh: trends and directions for the future.J Health Popul Nutr 2012, 30(1):1–11.

12. Khan MMH, Kraemer A, Khandoker A, Pruefer-Krämer L, Islam A: Trends insociodemographic and health-related indicators in Bangladesh, 1993-2007: will inequities persist? Bull World Health Organ 2011, 89(8):583–592.

13. Deolalikar AB: Poverty and child malnutrition in Bangladesh. J Deve Soc2005, 21(1–2):55–90.

14. Das S, Hossain MZ, Islam MA: Predictors of child chronic malnutrition inBangladesh. Proc Pakistan Acad Sci 2008, 45(3):137–155.

15. Smith LC, Haddad L: Explaining child malnutrition in developing countries: across country analysis. FCND Discussion Paper No. 60. Washington, DC:International Food Policy Research Institute (IFPRI); 1999.

16. Khanam R, Nghiem HS, Rahman MM: The impact of childhoodmalnutrition on schooling: evidence from Bangladesh. J Biosoc Sci 2011,43(4):437–451.

17. Gabriele A, Schettino F: Child malnutrition and mortality in developingcountries: evidence from a cross-country analysis. Munich: MPRA paper no.3132; 2007.

18. Hien NN, Kam S: Nutritional status and the characteristics related tomalnutrition in children under five years of age in Nghean, Vietnam.J Prev Med Publ Health 2008, 41(4):232–240.

19. Ojiako IA, Manyong VM, Ikpi AE: Determinants of nutritional status ofpreschool children from rural households in Kaduna and Kano states,Nigeria. Pakistan J Nutr 2009, 8(9):1497–1505.

20. Martin LG, Trussell J, Salvail FR, Shah NM: Covariates of child mortality inthe Philippines, Indonesia, and Pakistan: an analysis based on hazardmodels. Popul Stud 1983, 37(3):417–432.

21. Desai S, Alva S: Maternal education and child health: is there a strongcausal relationship? Demography 1998, 35(1):71–81.

22. Doak C, Adair L, Bentley M, Fengying Z, Popkin B: The underweight/overweight household: an exploration of household socio-demographicand dietary factors in China. Publ Health Nutr 2002, 5(1A):215–221.

23. Armstrong J, Dorosty AR, Reilly JJ, Team CHI, Emmett PM: Coexistence ofsocial inequalities in undernutrition and obesity in preschool children:population based cross sectional study. Arch Dis Child 2003, 88(8):671–675.

24. Thang NM, Popkin BM: In an era of economic growth, is inequity holdingback reduction in child malnutrition in Vietnam? Asian Pac J Clin Nutr2003, 12(4):405–410.

25. Zere E, McIntyre D: Inequities in under-five child malnutrition in SouthAfrica. Int J Equity Health 2003, 2(1):7.

26. Giashuddin MS, Kabir M, Hasan M: Economic disparity and child nutritionin Bangladesh. Indian J Pediatr 2005, 72(6):481–487.

27. Mashal T, Takano T, Nakamura K, Kizuki M, Hemat S, Watanabe M, Seino K:Factors associated with the health and nutritional status of children under5 years of age in Afghanistan: family behavior related to women and pastexperience of war-related hardships. BMC Publ Health 2008, 8:301.

28. Siddiqi MNA, Haque MN, Goni MA: Malnutrition of under-five children:evidence from Bangladesh. Asian J Med Sci 2011, 2:113–119.

29. Wamani H, Tylleskar T, Astrom AN, Tumwine JK, Peterson S: Mothers'education but not fathers' education, household assets or landownership is the best predictor of child health inequalities in ruralUganda. Int J Equity Health 2004, 3:9.

30. Fotso J-C: Urban–rural differentials in child malnutrition: trends andsocioeconomic correlates in sub-Saharan Africa. Health Place 2007, 13:205–223.

31. Roy NC: Use of mid-upper arm circumference for evaluation of nutritionalstatus of children and for identification of high-risk groups for malnutritionin rural Bangladesh. J Health Popul Nutr 2000, 18(3):171–180.

32. Chatters LM: Religion and health: public health research and practice.Annu Rev Publ Health 2000, 21:335–367.

33. Murray CJ, Lopez AD: Global mortality, disability, and the contribution of riskfactors: global burden of disease study. Lancet 1997, 349(9063):1436–1442.

34. Mueller O, Garenne M, Kouyate B, Becher H: The association betweenprotein-energy malnutrition, Malaria morbidity and all-cause mortality inwest African children. Trop Med Inter Health 2003, 8(6):507–511.

35. Winkelmann R, Zimmermann KF: Count data models for demographicdata. Math Popul Stud 1994, 48(3):205–221.

36. Wang W, Famoye F: Modeling household fertility decisions withgeneralized poisson regression. J Popul Econ 1997, 10:273–283.

37. Famoye F, Wulu JT Jr, Singh KP: On the generalized Poisson regressionmodel with an application to accident data. J Data Sci 2004, 2:287–295.

38. Xie F-C, Wei B-C: Influence analysis for count data based on generalizedpoisson regression models. Statistics 2010, 44(4):341–360.

39. National Institute of Population Research and Training (NIPORT), Mitra andAssociates, and Macro International: Bangladesh demographic and healthsurvey 2007. Dhaka, Bangladesh and Maryland, USA: NIPORT, Mitra andAssociates, and Macro International; 2009.

40. Waterlow JC, Buzina R, Keller W, Lane JM, Nichaman MZ, Tanner JM: Thepresentation and use of height and weight data for comparing thenutritional status of groups of children under the age of 10 years.Bull World Health Organ 1977, 55(4):489–498.

41. Bhagowalia P, Chen SE, Masters WA: Effects and determinants of mildunderweight among preschool children across countries and over time.Econ Hum Biol 2011, 9(1):66–77.

42. Consul PC, Jain GC: A generalization of the poisson distribution.Technometrics 1973, 15(4):791–799.

43. Shoukri MM: Estimation problems for some generalized discrete distributions.Ph.D. thesis. Calgary: University of Calgary; 1980.

44. Rayhan MI, Khan MSH: Factors causing malnutrition among under fivechildren in Bangladesh. Pakistan J Nutr 2006, 5(6):558–562.

45. Hong R: Effect of economic inequality on chronic childhoodundernutrition in Ghana. Publ Health Nutr 2007, 10(4):371–378.

46. Khan MMH, Kabir A, Khandoker A, Kabir M: Inequalities in fertility andmortality related indicators among ever married women in Bangladesh.J Stat Stud 2011, 30:15–26.

47. Frost MB, Forste R, Haas DW: Maternal education and child nutritionalstatus in Bolivia: finding the links. Soc Sci Med 2005, 60(2):395–407.

48. Hobcraft JN: Women’s education, child welfare and child survival: areview of the evidence. Health Trans Rev 1993, 3(2):159–173.

49. Frongillo EA Jr, De Onis M, Hanson KMP: Socioeconomic and demographicfactors are associated with worldwide patterns of stunting and wastingof children. J Nutr 1997, 127(12):2302–2309.

50. Larrea C, Kawachi I: Does economic inequality affect child malnutrition?the case of Ecuador. Soc Sci Med 2005, 60(1):165–178.

51. Ross NA, Wolfson MC, Dunn JR, Berthelot J-M, Kaplan GA, Lynch JW:Relation between income inequality and mortality in Canada and in theUnited States: cross sectional assessment using census data and vitalstatistics. BMJ 2000, 320:898–902.

doi:10.1186/1471-2458-13-11Cite this article as: Islam et al.: Predictors of the number of under-fivemalnourished children in Bangladesh: application of the generalizedpoisson regression model. BMC Public Health 2013 13:11.